038_Eugenio Picchi & Carol Peters & Elisabetta Marinai-The Pisa Lexicographic Workstation_ the Bi

8/18/2019 038_Eugenio Picchi & Carol Peters & Elisabetta Marinai-The Pisa Lexicographic Workstation_ the Bi

1/9

Eugenio Picchi(Istituto di Linguistica Computazionale, CNR, Pisa, ltaly),Carol Peters(Istituto di Elaborazione della Informazione, CNR, Pisa, ltaly)Elisabetta Marinai(ACQUILEX Project, Istituto di Linguistica Computazionale, CNR,Pisa, ltaly)

The Pisa LexicographicWorkstation: The BilingualComponents

ABSTRACT: The main components of the Pisa Lexicographic Workstation

are a full text retrieval system and a lexical database system; each system

Incorporates procedures that have been Implemented to meet the spe-

cific needs of the lexicographer. The paper describes the recent tailoring

ofexistlng modules and the development ofnew ones wfth bilingual lexi- cographyln mind. The alm Is to provide a flexible, userfriendlysystem that

can be employed In all stages of dictionary compilation, from the acquisi-

tion of citation material to the formatting of the entry for printing.

1. Introduction

For some time now, a lexicographic workstation has been under development at the

"Istituto di Linguistica Computazionale" in Pisa. The workstat ion provides a series oftools designed specifically for linguistic and lexicographic text processing tasks which

can be used by the lexicographer to assist him/her in the various activities involved in

the creation and revision of dictionaries. The main components of the workstation are the

DBT (Data Base Testuale), a full text retrieval system that has been developed to query

and analyse all kinds of texts and textual corpora (Picchi 1991), and a lexical database

system that has been implemented to handle dictionary acquisition and processing acti

vities; a morphological procedure is associated with the text and dictionary query sys

tems. The lexicographer can use these two systems, the DBT and the LDB, to interrogate

on-line text archives and electronic dictionaries and retrieve and extract reference andcitation material. The core module of the system is a procedure for on-line dictionary

editing which includes functions for windowing into and copying data from the diction

ary and text archives, and is integrated with a structured indexing procedure that can be

used to query the dictionary in compilation in order to check the regularity and consist

ency of the input. The present paper describes the recent specialization of existing mo

dules and the development of new ones to meet the specific needs of the bilingual

lexicographer.


2/9

278 EURALEX '92 - PROCEEDINGS

2 . The Bilingual Components

The bilingual modules consist of an on-line bilingual entry editor, a bilingual lexical

database query system, and a system for the automatic creation and retrieval of parallel

concordances from bilingual text archives (DBT-Synchro). As the languages which are

currently being considered for the bilingual components are Italian and English, we have

also added a morphological procedure for English to complement the program already

implemented for Italian. Figure 1 shows the global structure of the integrated mo no-/ bi -

lingual lexicographic workstation.

Figure 1 Integrated Monolingual and Bilingual Lexicographic Workstation

2.1. Bil ingual Dictionary Editor

The bilingual editor is a specialized version of the on-line editor developed for mono

lingual dictionary compilation, providing functions to assist the bilingual lexicographer

in creating or revising lexical entries. When the editor environment is entered, the user is

presented with a basic entry template. This schema represents explicitly, in tagged fields,

all the lexical information (phonetic , syntactic and semantic) which will be contained in

the printed dictionary. The default entry template has been designed on the basis of

experience acquired during the study of a standardized representation structure for dic-


3/9

Picchi et. al : The Pisa Lexi cogr aphi c Workstation 279

tionary entries in the ACQUILEX project (see Calzolari et a l. 1990). The use of a uniform

representation language ensures compatibility between entries, permits the exchange of

data between different projects in a common format, and facilitates analyses over different dictionaries

2

. However, the lexicographer is free to modify the structure of the tem

plate proposed by the system to meet the particular characteristics of the dictionary or

lexicon on which he is working.

The source language lexicographer, whose task it is to analyse the headword and

make a first proposal of sense divisions and example material, compiles his entries filling

in the appropriate data fields; they can then be saved on file or printed out and passed

on to the target language compiler who is responsible for transfer into L2 but who may

also propose changes to the original analysis on the basis of L2 dependent factors. The

final synthesis of the entry will be the result of the joint efforts of both lexicographers. Theresults of each stage can be saved separately so that a trace of the development of the

entry, through the several stages of analysis, transfer and synthesis, is maintained. At any

moment, the lexicographers can access both monolingual and bilingual text corpora and

LDBs from within the editor environment, which includes functions to window into,

query and copy information from these reference archives. The structure of the entry and

the main functions available for bilingual dictionary editing can be seen in Figure 2

which shows a screen dump from a work session in which the Help function has been

invoked.

E. Picchi - Bilingual Dictionary Editor - Italian/English CARRO

Hdwd 1 carro = Help IPron 2 ['karro] ' T ' Previous field

PoS 4 sm • 1 ' Next field

NPoS 5 N ' " ' Duplicate field

SenNo 6 (a) • C Change field code

Trans 7 cart, wagon; *D' Delete field

SICon 8 per carnevale •E' Edit field

Trans 9 float; •I' Insert new field

Ex 10 mettere il =**= avanti ai buoi 'J' Join two fields

SemCd 11 fig •R' Restore a fieldExTr 12 to put the cart before the horse. 'S' Split text field

SenNo 13 (b) ALT+F8 Restore entrySubCd 14 Astron ALT+F9 Copy entry

Ex 15 il Gran/Piccolo C=**= ALT+F10 Delete entry

ExTr 16 the Great/Little Bear. F3 Quit entry Mwd 17 =**= armato F4 Save entry

SubCd 18 MiI F5 Previous entry

MwdTr 19 tank; F6 Next entry

Mwd 20 =**= attrezzi F7 Call MLDB

SubCd 21 Aut F8 Call DBT

MwdTr 22 breakdown van; F9 Call DBTSynchro

i for more = = = = =

Select Function»

Figure 2: Bilingual Dictionary Editor

During work session, parts of the dictionary being compiled can be extracted and

printed or saved in separate files. In this way, particular subsets of the dictionary can be

treated independently: a) a given subset of the lexicon; b) a given subset of the data fields;

c) only entries satisfying certain constraints. These facilities can be adopted when it is


4/9


desired to have certain data fields compiled separately (e.g. the phonetic transcriptions)

or to make certain consistency checks throughout the dictionary (e.g. for semantic label

attribution).The compiled lexical entries can be input to the parsing and indexing procedures of

the bilingual lexical database system, and can then be checked interactively by the lexi

cographer using the LDB query system (see 2.3 below). Another procedure that can be

used on the entries in LDB form to check the consistency of the data is the Reverse

procedure, which operates on the data contained in the translation field of each entry in

order to recognize certain incongruities between the two bilingual data sets. The proce

dure identifies (i) all cases where translations on one side of the dictionary are not listed

as headwords on the other, and (ii) all cases where words, appearing as translat ions of a

lemma on one side of the dictionary, when listed as headwords on the other side do notgive the lemma as one of their possible translations. There are a number of reasons why

symmetry is not always desirable but the bilingual lexicographer should have access to

such information during the construction of the dictionary. He can then decide whether

an omission is deliberate or due to an oversight, and whether to make amends or not. The

Reverse procedure can also be executed in a similar way on words appearing in the

Example Translation held. In each case, the procedure is run automatically and the

results are printed out for verification by the lexicographer.

The completed lexical entries can be printed out in various formats, or used as input

for photocomposition systems which will produce the final version of the dictionary.

22. Morphologica l Procedure for EngUsh

Our morphological system consists of a language independent set of procedures which

operate on a suitably encoded description of a language in order to recognise and pro

duce words in that language. We adopt a two-level approach rather than a generative

one3

. The language description is formulated in two files: a lexicon file containing a list

of base lemmas with associated morphosyntactic information and an inflection code; a

rule file containing the rules which specify the correspondences between underlyinglexical items and surface forms. The program is reversible: the same lexicon and set of

rules is used for recognition and for generation. Our guiding principle has been effi

ciency and convenience of implementation.

This module has already been implemented for Italian (see Picchi and Calzolari 1990).

For the English version we have derived our list of lemmas from the headwords of two

computerized dictionaries. Each lexical entry has associated morphosyntactic informa

tion. A code has been assigned (semi)automatically to each lemma and is used to invoke

the rule which determines its inflection. Rules have been written to cover all regular

inflections and those irregular inflections which can be grouped together into classes.Highly irregular inflections are encoded singly. Information on irregular morphology

and other phenomena such as gemination has been extracted from the computerized

dictionaries; we use orthographic rather than phonological information as this informa

tion is more conveniently derived from our dictionaries.

The morphological tool includes an on-line display and editor which can be used to

view the generation of the word forms for any lemma in the lexicon and to add to or

correct either the inflectional or morphosyntactic codes if necessary. The lexicographer


5/9

PiccN et. al : The Pisa Lexicographic Workstation 281

с п use the generator when querying text or dictionary archives to expand any given

lemma by producing the set of all its forms; the whole paradigm for the lemma can then

be searched by entering a single command. The complete morphological procedure

(English/Italian analyzers and generators) is also an essential component of the bilingual

text retrieval system.

2 3 . T he Bilingual Lexical Database Query System

A major component of the set of tools implemented with the bilingual lexicographer in

mind is the bilingual lexical database system. The various stages in the design and

development of this system have already been described elsewhere and the bilingual

LDB now forms part of a Multilingual Lexical Database System (MLDB) that is beingimplemented in the context of the ACQUlLEX project (see Marinai et al. 1990).

The bilingual LDB query system provides dynamic search procedures which permit

the user to navigate through the dictionary data and within the different fields of the

entry in order to access and retrieve information of interest in whatever part of the

dictionary it is stored, specifying the language on which the query is to operate. In this

way, much information which is normally '^idden" in the printed dictionary can be

accessed and exploited. The query system supplies a series of functions which can be

used to look up lexical items or combinations of items. The user can search given items

or character strings, define search functions in which items or character strings are combined by AND, OR and NOT operators, retrieve all entries satisfying the search condi

tions, display, print or store on file all or a selected part of the results, and define restr ic

tion rules on the results of a previous search. Searches are made on an attribute-value

basis and the results are given for each field in which the item is found. The lexicographer

can use the query system on existing dictionaries maintained by the system or on the

dictionary under construction, which can be input to the MLDB parsing and indexing

procedures and structured as an on-line database.

LDB (E. Picchi) Collins Bilingual Dictionary (Normalized) VItem : {I} CARRO Frequency : 9

1) C H A R I O T {Hdwd} chariot {PoS} n {NPoS} N {Trans} cocchio, carro.2) D U S T C A R T {Hdwd} dustcart {PoS} n {NPoS} N {Trans} carro dellanettezza urbana or delle immondizie.3) F L O A T gen {Trans} galleggiante {m}; {SI} cork {Trans} sughero;{SI} in procession {Trans} carro; {SI} sum of money {Trans} somma.4) H E A R S E {Hdwd} hearse {PoS} n {NPoS} N {Trans} carro funebre.5) H O R S E B O X {Hdwd} horsebox {PoS} n {NPoS} N {Trans} carro orfurgone {m} per il trasporto dei cavalli.6) T A N K {Ex} fuel =**= {ExTr} serbatoio del carburante. {SenNo} (b){SubCd} MiI {Trans} carro armato.7) T R U C K {PoS} n {NPoS} N {SenNo}


6/9


Figure 3 gives just one example of how the LDB can be used to acquire information on

a given item that is scattered throughout the dictionary and otherwise inaccessible. The

Italian lemma CARRO, shown in Figure 2, was searched throughout the LDB which hasbeen constructed for the Collins Concise English/Italian dictionary; the figure displays

all those entries in which CARRO was found in the Translation field.

A procedure has also been implemented to permit semi-automatic mapping between

monolingual and bilingual LDBs in the workstation. This procedure provides the biling

ual lexicographer with a useful tool that permits him to examine and compare the lexical

information given for the same item in different source dictionaries (for a full description

see Marinai et aL , forthcoming).

2.4. Bilingual Text Retrieval

The most recent component to be added to the Lexicographic Workstation is a system for

the automatic construction and retrieval of parallel contexts from bilingual text archives.

The importance of large language reference corpora in monolingual lexicography is

widely acknowledged and it has also been asserted that such corpora are useful for the

bilingual lexicographer (see, for example , Atkins 1990). However, interest is now grow

ing in the potential of bilingual corpora as valuable sources of documented evidence on

the relationships between two languages. We feel that ideally the bilingual lexicographer

should have access to both sources. The monolingual corpus will be used during the

analysis stage helping to achieve an accurate first breakdown of a given headword into

senses and to retrieve valid examples of usage and collocations; the bilingual corpus will

be employed in transfer helping to find appropriate real world translations for each

usage of the Ll word suggested by the source language compiler, reflecting parallelisms

or differences in sense divisions. For example, a particular sense of a word in Ll , estab

lished on the basis of corpus evidence, may well have no single equivalent or set of

equivalents in L2 to cover the full scope of L l . The bilingual corpus can be used to study

carefully how each use is rendered in order to group the TL equivalents. It may well

provide evidence which suggests an adjustment of an Ll sense division to meet the

demands o f L2.

Of course, this implies the availability of a high quality, sufficiently representative

bilingual corpus. However, the construction of a text corpus implies a considerable in

vestment of time and resources. Before any decisions are taken, the criteria to be adopted

when assembling corpus material must be carefully evaluated. In particular, each type of

text sample must be labelled, as source and target texts reveal different uses of language;

it is claimed that a translation is never a true representation of the language in which it is

written but rather reflects the relationships between the language of the target and that

of the source. At Pisa, we have assembled a sample set of bilingual texts, selected to cover

a number of different language varieties, ranging from scientific papers to poetry, from

university text books to magazine articles. This set of texts was collected in the first place

to provide a test-bed for our bilingual retrieval system but should also supply useful data

to assist us in the definition of design criteria, which can then be used in a subsequent

extension of these archives.

So far, mos t of the bilingual concordancing tools implemented for the lexicographer

use statistically based programs to align the texts at the sentence level. But, as stated by


7/9

Picchi et. al : The Pisa Lexicographic Workstation 283

Church and Gale (1991), such sentence based concordance programs are not very good

at showing what is not already known as the user is requested to supply the program

with both a SL word and a TL candidate translation. These authors aiso describe aword-based concordance tool in which the possible translations for a given word are

discovered from the corpus by the program, using a pnxx>mputed index indicating

which words in one language correspond to which words in the other. We have adopted

a different approach which depends on the use of external evidence (extracted from a

computerized bilingual dictionary) to create direct links between parallel texts on the

basis of translation equivalents. In this way, we exploit already known information (dic

tionary translations) to access "unknown" information (real world TL renderings).

A preliminary version of our system, known as DBT-Synchro, is described in Marinai

et al . (1991). It operates in two distinct steps. In the first, sets of bilingual texts are"synchronized" using morphological procedures and a bilingual electronic dictionary

(based on the Collins LDB). Each word form in the text taken as the source is input to the

morphological analyzer for Ll in order to identify its base lemma, which is then searched

in the bilingual LDB. AIl translations read for this lemma are input to the morphological

generator for L2 which produces all possible forms and these are then searched over the

relevant search zone in the target text. When one of the translation equivalent forms are

found in the L2 text, a link is created between this form and its equivalent in the Ll text.

These links are then stored with the texts in the bilingual archives to be used by the query

system for the on-line construction of parallel contexts. 'Wrong" links between falsely

recognized translation equivalents which disturb context calculation are identified and

eliminated by the query system, which then recalculates the parallel contexts on the basis

of those links recognised as valid.

D.B.T. (Picchi) Testini di prova della sincronizzazione V{E)CHANGE & {E)COLOUR

1 {E} September days continue, followed by those splendid October days marked by golden sunset and skies which ch*ng* colour from green to goldas in the Venetian paintings of Cima da Conegliano and Titian.E-impressi.63

{1} Ie giornate di sole della fine di settembre, e Ie splendideottobrate dai tramonti dorati, dai ci#tl che trascolorano dal verde, all'oro, come quelli che si trovano nella pittura veneta da Cima diI-impressi.65

2 {E} the melancholy signs that the season is hurrying to an end.The corn begins to ch*ng* colour in the fields. The leaves are partlygreen and partly the dry dirty gray colour E-impressi.l34

{1} suoi segnali melanconici di stagione che corre verso Ia fine. Le piante del granoturco cominciano a c*mbi*r* color* nel campo. Le fogliesono mezzo verdi e mezzo secche, di un grigio sporco in I-impressi.l31

3 {E} that winter is now just around the corner. The leaves on all the

trees have started to Chang* colour. Everything begins to turn yellow orred. Though it seems like a time of celebration E-impressi.l67

{1} e che 1' inverno è ormai dietro Ia porta. Le foglie di tutti glialberi hanno cominciato a mutar* color*. Tutto inizia a ingiallire o arosseggiare,Pare un momento di festa e di trionfo I-impressi.l61

Scelta N.Contesto> Fl for help

Figure 4: Querying the Bilingual Text Corpus


8/9


The archives are considered to be symmetric; either of the two languages can be

selected as Ll . The lexicographer can either search for single word forms or, using the

morphological generator, for all the forms of a given lemma. For each Ll word or combination of words queried by the user, the parallel L2 contexts are constructed and dis

played on the screen. The word(s) for which the contexts are being created are high

lighted and where a direct link for the Ll form(s) being searched exists, the L2 matched

word(s) will be highlighted in the same colour. Otherwise, the two directly linked forms

which are closest to the point calculated as the middle of the L2 context will be evidenced

in a different colour, as indicators of the position in the L2 context of the translation

equivalent being searched. Other words which have been linked in the paired contexts

can be optionally evidenced. Bilingual concordances of interest can be printed out or

saved in a separate file for future reference. Figure 4 gives an extract of the resultsobtained from our test set of texts for a query requesting parallel contexts for the cooc

currence of CHANGE and COLOUR.

3. Final Remarks

We have given a brief description of the main computational tools which have been

implemented in the Pisa Lexicographic Workstation in order to facilitate the task of the

bilingual lexicographer. The entire system is implemented on personal computers run-

ning the MS/DOS operating system and is intended to run on a Local Area Network sothat a team of lexicographers can work in unison, using the same tools and accessing the

same reference data. At the same time, the procedures are easily transportable onto

smaller desk-top systems for the lexicographer working at home. The system is menu-

driven and context sensitive Helps are accessible atall times during a query session. Our

main consideration has been to provide tools which are not only efficient but also user-

friendly.

Endnotes

1 ACQUlLEX is an ESPRTT Basic Research Action which is developing techniques and

methodologies for utilising both monoUngual and biUngual machine-readable dictionarysources to construct lexical components for NLP systems.

2 The study made for ACQUTLEX has also been used by the Text Encoding hùtiative group studying dictionary representation as part of the general TEI programme to provide guidelinesand standards for the representation and exchange of texts in machine-readable form usingthe SGML mark-up kmguage (see Ide et al. 1991). We wiU consider incorporating the final recommendations of this group into our actual data model.

3 For a description of the two-level model versus generative phonology, see Antworth (1990,Introduction).


9/9

PiccN et . al: The Pisa Lexicographic Workstation 285

Bibliography

ANTWORTH, E.L. (1990): PC-KlMMO: A Two-level Processor for Morphological Analysis. Occasional Publications in Academic Computing. No.l6, Summer mstitute of Linguistics, Dallas.

ATKlNS, B. (1990): "Corpus Lexicography: The Bilingual Dimension", m: Computational Lexicology and Lexicography. Special Issue dedicated to Bernard Quemada. I. Ed. by L. Cignoni andC Peters. Linguistica Computazionale, VoI VI.

CALZOLARI, N., PETERS, C., ROVENTTM, A. (1990): Computational Model of the Dictionary

Entry: Preliminary Report. ACQUILEX, Esprit BRA 3030. Six Month Deliverable. ILC-ACQ-I-

90 ,Pisa.

CHURCH, K., GALE, W. (1991): "Concordances for Parallel Text", hi: Using Corpora. Proceedings

of the 7th Annual Conference of the Centre for the New Oxford EngUsh Dictionary and Text

Research, Oxford, UK.IDE, N., VERONIS, J., WARWICK-ARMSTRONG, S., CALZOLARI, N. (1991): Principles for En

coding Machine Readable Dictionaries, TEIWP, A15W6.

MARTNAI, E., PETERS, C., PICCHI, E. (1990): The Pisa Multilingual Lexical Database System.Esprit BRA 3030. Twelve Month DeUverable. uX-ACQ-2-90, Pisa.

MARJNAI, E., PETERS, C., PICCHI, E.: "A prototype system for the semi-automatic sense linking

and merging of mono- and bilingual LDBs". m: Research in Humanities Computing. Ed. by

N.Ide and S. Hockey. OUP, Oxford, (forthcoming).

MARTNAI, E., PETERS, C., PICCHI, E (1991): bilingual Reference Corpora: A System for Parallel

Text Retrieval". In: Using Corpora. Proceedings of the 7th Annual Conference of the Centre for

the New Oxford EngUsh Dictionary and Text Research, Oxford, UK.

PICCHI, E. (1991): "D.B.T.: A Textual Data Base System". In: Computational Lexicology and Lexicography. Special Issue dedicated to Bernard Quemada. II. Ed. by L. Cignoni and C Peters.Linguistica Computazionale, VoI VII.

PICCHI, E., CALZOLARI, N. (1990): "Pisa Linguistic Database". In: Literary and Linguistic Computing 1988, Proceedings of ALLC Fifteenth International Conference. Ed. by Y.Choueka.Champion-Slatkine, Paris

Date post:	07-Jul-2018
Category:	Documents
Upload:	escarlata-ohara
View:	216 times
Download:	0 times

038_Eugenio Picchi & Carol Peters & Elisabetta Marinai-The Pisa Lexicographic Workstation_ the Bi

Documents